Process_Description:
Multi-date Landsat 7 Enhanced Thematic Mapper (TM) satellite imagery were used to
generate this land use classification for 2001. Cloud coverage dictated date selection
to the greatest extent, although imagery were chosen to represent spring, summer, and
fall growing conditions. Before the supervised classification was run, clouded and urban
areas areas were removed from the imagery. Urban areas were defined using 2000 TIGER
cenus data. Clouded areas were classified in a second stage, using an unsupervised
classification with the remaining unclouded dates. Field data for crops were obtained
using Farm Service Agency (FSA) reporting records from 2001. Spectral signatures
for cropland were collected by identifying homogeneous areas defined on FSA reporting
records. The National Wetlands Inventory (NWI) was used to locate wetland areaa to
collect spectral signatures. 1993 and 1999 Digital Orthohoto Quarter Quadrangles were
used to locate signature areas for riparian and forested areas, roads, and
non agricultural features. Multiple spectral signatures for each class were extracted
and evaluated for continuity. Unrepresentative signatures were deleted. The remaining
signatures were then used in a supervised classification on the multi-date imagery.
After the classification algorithm was run, classes were evaluated for accuracy. A
'cluster busting' technique was applied to compensate for mixed classes. Irrigation
information was obtained from FSA reporting records, Nebraska Department of Natural
Resources, Public Power Districts, and Nebraska Natural Resource Districts. Center
pivots were on-screen digitized using the 2001 Landsat ETM data. Once the land cover
classification was complete, a 3X3 majority filter was appied to agricultural classes
to minimize the remaining mixed pixels. The cell size for the data is 93.48 feet and
there are 25 different land use/land cover classes. The overall accuracy of the
land cover classification was calcuated at 82.7%